Project Summary Understanding the conformational dynamics of proteins and their binding partners are crucial to predicting and designing their function. As computer hardware and software becomes ever more efficient, simulation- based methods will play increasingly important roles in molecular design. Markov State Models (MSMs), which describe conformational dynamics as a network of transitions between metastable states, can be used as simulation-based platform for the prediction and design of multiple sequences?for small perturbations that preserve state definitions, mutational effects can be inferred by estimating changes in transition rates?but new methods must be developed to do this efficiently. We will address this challenge by developing new methods to efficiently sample MSMs for multiple sequences, and then apply this technology to predict and design binding affinities and rates of peptidomimetics, an area that will have widespread benefits to human health. Our first specific aim is to develop two analytic tools facilitating the efficient estimation of MSMs for multiple sequences: (1) surprisal-based adaptive sampling, which uses a relative entropy metric for two or more MSMs to prioritize sampling of states that most efficiently decrease the uncertainty in the models, and (2) maximum- caliber approaches for inferring changes in MSM transition rates directly from changes in state populations. These methods will be tested against changes in stabilities and folding rates measured for a corpus of well- studied mini-proteins with available trajectory data. Our second specific aim is to apply this technology to predict binding affinities, pathways and rates for peptidomimetic ligands of MDM2, a well-studied protein-peptide binding system and important cancer target. We will build MSMs of apo-MDM2 to explore the role of the N-terminal lid region in ligand binding, and the utility of MSM-derived receptor ensembles for computational drug design. We will then construct an MSM of p53 binding to MDM2, and use it as a starting point for building multi-ensemble MSMs of ligand binding for series of related small-molecules, peptides, and spiroligomer peptidomimetics, with the goal of achieving efficient estimates of affinities as well as binding on- and off-rates. Our third specific aim, a collaboration with the David Baker lab, is to use MSM methods to screen and improve de novo designed protein binders of LapG, a new route to disperse bacterial biofilms, a major source of antibiotic resistance. Toward this end, we screen the binding properties of about 100 top-ranked designs and choose around a dozen for expression, purification and assaying for binding. If successful, we will have avoided the need for time-consuming yeast display experiments, moving a step closer to a self-contained computational pipeline for generating custom protein binding interfaces, a potentially transformative tool in biology and medicine. We will evaluate methods for ?in silico affinity maturation? against experimental data for a site-saturated library of all possible single-point mutations measured for our top-binding candidate.